Quantification of brown tide algae using EGAF coding and DFF feature fusion based on LED-induced fluorescence spectroscopy

The distribution of brown tide algae in the water body can reflect the change of water quality, which is the key to the prevention and treatment of brown tide. Therefore, it is crucial to monitor the concentration of brown tide algae. In this work, an improved gramian angular field (GAF) for fluorescence spectroscopy coding combined with attention mechanism and convolutional neural network (CNN) for brown tide algae concentration prediction was proposed. First, an improved enhanced gramian angular field (EGAF) was used to encode one-dimensional data into two-dimensional image data to provide the hidden features and the correlation between elements under the two-dimensional representation for the one-dimensional spectra; second, a dual feature fusion (DFF) network designed based on the attention mechanism was used to perform feature fusion on the EGAF maps; and finally, CNN was utilized to complete the extraction of feature information and the prediction of algae concentration. The experimental results show that the EGAF-DFF-CNN model proposed in this work can effectively predict the algae concentration with high accuracy.